Swiss researchers found that large models like GPT-4 can infer users' private information such as age, gender, and income from Reddit posts. GPT-4 achieved over 60% accuracy in predicting this information on the Personal Reddit dataset, especially for gender and birthplace. As model scale increases, GPT-4's accuracy in attribute inference improves significantly. The authors successfully extracted user information with a comparable accuracy to GPT-4 through simulated chatbot experiments. Experts point out that identifying and removing personal data from large models is nearly impossible, highlighting the need for privacy protection measures.